powerpc/powernv: Fix restore of SPRs upon wake up from hypervisor state loss
authorGautham R. Shenoy <ego@linux.vnet.ibm.com>
Wed, 7 Sep 2016 05:16:30 +0000 (10:46 +0530)
committerMichael Ellerman <mpe@ellerman.id.au>
Mon, 12 Sep 2016 02:45:50 +0000 (12:45 +1000)
pnv_wakeup_tb_loss() currently expects cr4 to be "eq" if the CPU is
waking up from a complete hypervisor state loss. Hence, it currently
restores the SPR contents only if cr4 is "eq".

However, after commit bcef83a00dc4 ("powerpc/powernv: Add platform
support for stop instruction"), on ISA v3.0 CPUs, the function
pnv_restore_hyp_resource() sets cr4 to contain the result of the
comparison between the state the CPU has woken up from and the first
deep stop state before calling pnv_wakeup_tb_loss().

Thus if the CPU woke up from a state that is deeper than the first
deep stop state, cr4 will have "gt" set and hence, pnv_wakeup_tb_loss()
will fail to restore the SPRs on waking up from such a state.

Fix the code in pnv_wakeup_tb_loss() to restore the SPR states when cr4
is "eq" or "gt".

Fixes: bcef83a00dc4 ("powerpc/powernv: Add platform support for stop instruction")
Signed-off-by: Gautham R. Shenoy <ego@linux.vnet.ibm.com>
Reviewed-by: Shreyas B. Prabhu <shreyasbp@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
arch/powerpc/kernel/idle_book3s.S

index 2265c63..bd739fe 100644 (file)
@@ -411,7 +411,7 @@ END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
  *
  * r13 - PACA
  * cr3 - gt if waking up with partial/complete hypervisor state loss
- * cr4 - eq if waking up from complete hypervisor state loss.
+ * cr4 - gt or eq if waking up from complete hypervisor state loss.
  */
 _GLOBAL(pnv_wakeup_tb_loss)
        ld      r1,PACAR1(r13)
@@ -453,7 +453,7 @@ lwarx_loop2:
         * At this stage
         * cr2 - eq if first thread to wakeup in core
         * cr3-  gt if waking up with partial/complete hypervisor state loss
-        * cr4 - eq if waking up from complete hypervisor state loss.
+        * cr4 - gt or eq if waking up from complete hypervisor state loss.
         */
 
        ori     r15,r15,PNV_CORE_IDLE_LOCK_BIT
@@ -481,7 +481,7 @@ first_thread_in_subcore:
         * If waking up from sleep, subcore state is not lost. Hence
         * skip subcore state restore
         */
-       bne     cr4,subcore_state_restored
+       blt     cr4,subcore_state_restored
 
        /* Restore per-subcore state */
        ld      r4,_SDR1(r1)
@@ -526,7 +526,7 @@ timebase_resync:
         * If waking up from sleep, per core state is not lost, skip to
         * clear_lock.
         */
-       bne     cr4,clear_lock
+       blt     cr4,clear_lock
 
        /*
         * First thread in the core to wake up and its waking up with
@@ -557,7 +557,7 @@ common_exit:
         * If waking up from sleep, hypervisor state is not lost. Hence
         * skip hypervisor state restore.
         */
-       bne     cr4,hypervisor_state_restored
+       blt     cr4,hypervisor_state_restored
 
        /* Waking up from winkle */